Multivariate Network Meta-Analysis for Predicting Treatment Effect from Surrogate Endpoints
*Sylwia Bujkiewicz, University of Leicester
Keywords: surrogate endpoints, multivariate network meta-analysis
In early regulatory decision making, evidence of effectiveness of new interventions may be limited to short-term surrogate endpoints. Candidate surrogate endpoints are not always perfect, and the level of association between the treatment effect on surrogate and final outcomes may vary between treatments. While multivariate meta-analysis (MVMA) can be used to model surrogate endpoints, it does not differentiate between the treatment options. Network meta-analysis (NMA) combines data from trials investigating heterogeneous treatment contrasts and has the advantage of estimating effects for all contrasts individually. We exploit this framework to model surrogate endpoints by the use of multivariate NMA (MVNMA). We extend the model developed by Achana et al. (BMC Med Res Meth 2014), which enables borrowing of strength across treatments and outcomes, by modeling the relationship between the treatment effects on the surrogate and final endpoint. We use the product normal formulation developed for the MVMA (Bujkiewicz et al. Stat Med 2013), which also was adopted to model multiple surrogate endpoints. The modeling techniques are investigated using an example from multiple sclerosis.